Very few growth moats will survive AI - Sandy Diao [Descript]

AI erases growth moats, leaving unfair distribution as the edge

Sandy Diao's growth career has been a search for one unfair distribution advantage at a time. At Pinterest, support tickets told her that users and businesses needed different products, and the sign-up fork she built became the ad platform. At Descript she found a channel that carried a third of revenue and a pricing change that failed within weeks. Her current argument is that AI erases most growth moats, leaving a natural, unfair distribution advantage as the scarce asset.

Takeaways

• Few growth moats survive AI. Diao says moats change with company size and stage, and distribution is moving toward generative engine optimisation and algorithmic video. The durable edge is an unfair, natural distribution advantage, not a channel any competitor can buy.
• Early growth rewards directional insight over precision. Diao calls her approach data inspired, meaning her teams look for directional insight rather than precision. At Descript a reverse-trial pricing change, backed by benchmarking and user surveys, halved conversion or worse in the exposure group, and the company reverted to its original pricing.
• Descript's affiliate program drove a third of revenue. Diao built it with a third-party tool and Stripe-based tracking. Creators showed the product inside real workflow videos, so viewers did not feel advertised to. Some affiliates bid on Descript's brand terms in Google Ads and were eventually banned.
• Pinterest's sign-up fork became its ad platform. Support tickets revealed that small and large marketers were already using Pinterest. Asking new accounts whether they were regular users or businesses forked businesses into their own profile and experience, which grew into Pinterest's advertising platform.
• Onboarding emails only worked once triggered. Diao copied the peer formula of five to ten education emails a day, then found fatigue set in by the seventh. Pinterest moved to action-triggered emails built around what users had or had not done inside the product.
• Growth teams should over-index on big swings. Teams that avoid big swings end up testing and micro-optimising toward a local minimum. Diao runs two budgets, one of time and one of money, and puts almost all of the time budget on big bets that can find power-law channels.

Building AI agents for growth work

• Diao is building a stack of AI agents to orchestrate work across growth teams.
• She uses Claude Code as her growth operating system and calls its Skills Creator her favourite tool, and is deploying the agents with the teams she advises to test whether the output is high quality and high value.